Graduation - 2024 - December (Open Access)

Permanent URI for this collection

https://scholar.sun.ac.za/handle/10019.1/125747

Browse

Browsing Graduation - 2024 - December (Open Access) by browse.metadata.advisor "Botha, Matthys, M."

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Localised Solution Methods for Efficient Analysis of Large Antenna Arrays
    (Stellenbosch University, 2024-12) Conradie, André Stephanus; Botha, Matthys, M.; Stellenbosch University. Faculty of Engineering. Dept. of Electrical & Electronic Engineering.
    Antenna arrays are widely used in modern wireless communication and remote sensing systems. The design and calibration of antenna arrays is complicated by mutual coupling effects which can distort the electromagnetic properties of individual elements. These effects are typically accounted for by computing embedded element patterns (EEPs), which are the radiation patterns of individual elements in their radiating environments. The method of moments (MoM) is well suited for these computations, however, solution costs grow quickly with increasing array size. The growing demand for rapid and precise solutions to problems of increasing sizes ensures that fast methods in the MoM remain an active area of research. This dissertation focuses on developing fast iterative localised solution methods for the MoM analysis of large arrays with disjoint elements. Localisation means that the current on each element is solved in turn, with the approximate inclusion of mutual coupling effects. This is done iteratively to converge to the MoM solution. New methods are presented, all exhibiting fast, controllable iterative convergence and comprehensive error control. Owing to their localised nature, these methods are ideally suited to parallelisation. The residual-driven iterative radius-based domain Green’s function method (RD-IRBDGFM) is formulated for arrays with identical, disjoint elements. The convergence rate is controlled through the size of local solution domains. Various methods to reduce the required number of iterations and the cost of each iteration are introduced. Fast iterative convergence is demonstrated for diverse arrays. The preconditioned residual-driven direct coupling technique (PRD-DCT) uses preconditioning to enable fast iterative convergence without the need for overlapping local solutions. Elements are also not required to be identical. A single-level nested cross approximation (NCA) scheme is incorporated to accelerate matrix operations. Faster runtimes than those achievable with the multilevel fast multipole method (MLFMM) in the leading commercial solver, FEKO, are demonstrated for very large arrays. The NCA-PRD-DCT extends the PRD-DCT by improving its runtime performance. This is achieved by introducing a multilevel NCA compression scheme and by leveraging the Woodbury identity to increase the efficiency of local solutions and preconditioner construction. Runtime improvements of more than an order of magnitude over FEKO’s MLFMM are demonstrated.